ssr: register merging not working when value used outside of the basic block #2
Labels
bug
Something isn't working
Comments
huettern
added a commit
that referenced
this issue
Aug 11, 2021
ssr setup pseudo instruction framework move ssr pseudo code to RISCVExpandPseudoInsts revert changes to isellowering wip: implemented ssr pop move SSR expansion to a seperate pass ssr pseudo expansion working ssr pseudo: cleanup debug ssr: add bound/stride setup instructions for 1,2,3,4D builtins and IR intrinsic for ssr setup WIP ssr: new setup order, add push ssr: intrinsic working, reverted order of setup cleanup comment ssr: pseudo push may load and store ssr: add pop intrinsic ssr: add enable/disable pseudo ssr: add enable/disable intrinsic ssr: move float reg selection to static function ssr: move expand pass to preEmit phase and replace push pop with target specific move ssr: back to pre reg alloc and fmv, add experimental register clobbering ssr: registers need defining or else, optimization fails to complete ssr: registers reserved in functions that use SSR ssr: cleanup change ssr register offset to word, not byte builtind pointer type to void clobber 3 ssr registers ssr: fix wrong register unpacking ssr: add test for register reserving ssr: switch encoding to original clang: intrinsics for bound stride setup ssr: first take on register merging test: fix encoding ssr: test for register merge ssr: optional register merging through command line argument ssr: mark region during register merge path by parsing CSR access ssr: add intrinsics for read/write/repetition setup ssr: move debug statement into macro ssr: remove debug ssr: mark ssr regs live in in all MBB between en/dis for usage in loops ssr: mark ssr regs live during merging in all blocks ssr: make live ins unique before exiting ssr: reduce liveins to only blocks containing push/pop ssr: fixing a duplicated entry for repetition pseudo #2
pulp-bot
pushed a commit
that referenced
this issue
Sep 4, 2021
ssr setup pseudo instruction framework move ssr pseudo code to RISCVExpandPseudoInsts revert changes to isellowering wip: implemented ssr pop move SSR expansion to a seperate pass ssr pseudo expansion working ssr pseudo: cleanup debug ssr: add bound/stride setup instructions for 1,2,3,4D builtins and IR intrinsic for ssr setup WIP ssr: new setup order, add push ssr: intrinsic working, reverted order of setup cleanup comment ssr: pseudo push may load and store ssr: add pop intrinsic ssr: add enable/disable pseudo ssr: add enable/disable intrinsic ssr: move float reg selection to static function ssr: move expand pass to preEmit phase and replace push pop with target specific move ssr: back to pre reg alloc and fmv, add experimental register clobbering ssr: registers need defining or else, optimization fails to complete ssr: registers reserved in functions that use SSR ssr: cleanup change ssr register offset to word, not byte builtind pointer type to void clobber 3 ssr registers ssr: fix wrong register unpacking ssr: add test for register reserving ssr: switch encoding to original clang: intrinsics for bound stride setup ssr: first take on register merging test: fix encoding ssr: test for register merge ssr: optional register merging through command line argument ssr: mark region during register merge path by parsing CSR access ssr: add intrinsics for read/write/repetition setup ssr: move debug statement into macro ssr: remove debug ssr: mark ssr regs live in in all MBB between en/dis for usage in loops ssr: mark ssr regs live during merging in all blocks ssr: make live ins unique before exiting ssr: reduce liveins to only blocks containing push/pop ssr: fixing a duplicated entry for repetition pseudo #2
SamuelRiedel
pushed a commit
that referenced
this issue
Nov 1, 2022
We experienced some deadlocks when we used multiple threads for logging using `scan-builds` intercept-build tool when we used multiple threads by e.g. logging `make -j16` ``` (gdb) bt #0 0x00007f2bb3aff110 in __lll_lock_wait () from /lib/x86_64-linux-gnu/libpthread.so.0 #1 0x00007f2bb3af70a3 in pthread_mutex_lock () from /lib/x86_64-linux-gnu/libpthread.so.0 #2 0x00007f2bb3d152e4 in ?? () #3 0x00007ffcc5f0cc80 in ?? () #4 0x00007f2bb3d2bf5b in ?? () from /lib64/ld-linux-x86-64.so.2 #5 0x00007f2bb3b5da27 in ?? () from /lib/x86_64-linux-gnu/libc.so.6 #6 0x00007f2bb3b5dbe0 in exit () from /lib/x86_64-linux-gnu/libc.so.6 #7 0x00007f2bb3d144ee in ?? () #8 0x746e692f706d742f in ?? () #9 0x692d747065637265 in ?? () #10 0x2f653631326b3034 in ?? () #11 0x646d632e35353532 in ?? () #12 0x0000000000000000 in ?? () ``` I think the gcc's exit call caused the injected `libear.so` to be unloaded by the `ld`, which in turn called the `void on_unload() __attribute__((destructor))`. That tried to acquire an already locked mutex which was left locked in the `bear_report_call()` call, that probably encountered some error and returned early when it forgot to unlock the mutex. All of these are speculation since from the backtrace I could not verify if frames 2 and 3 are in fact corresponding to the `libear.so` module. But I think it's a fairly safe bet. So, hereby I'm releasing the held mutex on *all paths*, even if some failure happens. PS: I would use lock_guards, but it's C. Reviewed-by: NoQ Differential Revision: https://reviews.llvm.org/D118439 (cherry picked from commit d919d02)
SamuelRiedel
pushed a commit
that referenced
this issue
Nov 1, 2022
ssr setup pseudo instruction framework move ssr pseudo code to RISCVExpandPseudoInsts revert changes to isellowering wip: implemented ssr pop move SSR expansion to a seperate pass ssr pseudo expansion working ssr pseudo: cleanup debug ssr: add bound/stride setup instructions for 1,2,3,4D builtins and IR intrinsic for ssr setup WIP ssr: new setup order, add push ssr: intrinsic working, reverted order of setup cleanup comment ssr: pseudo push may load and store ssr: add pop intrinsic ssr: add enable/disable pseudo ssr: add enable/disable intrinsic ssr: move float reg selection to static function ssr: move expand pass to preEmit phase and replace push pop with target specific move ssr: back to pre reg alloc and fmv, add experimental register clobbering ssr: registers need defining or else, optimization fails to complete ssr: registers reserved in functions that use SSR ssr: cleanup change ssr register offset to word, not byte builtind pointer type to void clobber 3 ssr registers ssr: fix wrong register unpacking ssr: add test for register reserving ssr: switch encoding to original clang: intrinsics for bound stride setup ssr: first take on register merging test: fix encoding ssr: test for register merge ssr: optional register merging through command line argument ssr: mark region during register merge path by parsing CSR access ssr: add intrinsics for read/write/repetition setup ssr: move debug statement into macro ssr: remove debug ssr: mark ssr regs live in in all MBB between en/dis for usage in loops ssr: mark ssr regs live during merging in all blocks ssr: make live ins unique before exiting ssr: reduce liveins to only blocks containing push/pop ssr: fixing a duplicated entry for repetition pseudo #2 (cherry picked from commit 925e044)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
in case the use of an ssr_pop(0) is in a different BB, the register merging just eliminates the
popbut doesn't replace the actual use of e.g.ft0.this then leads to a crash of the compiler.
It can be mitigated with
-mllvm -ssr-noregmergebut that is not the idea of SSRsThe text was updated successfully, but these errors were encountered: